Analytical ecological epidemiology: exposure–response relations in spatially stratified time series

Abstract

AbstractAn important task of environmental research is the investigation of a possible causal relationship between exposure and the frequency of a biologic trait. Major industrial accidents provide examples where the exposure status of large populations may change considerably within relatively short time intervals of days or weeks (e.g. Seveso herbicide plant explosion, Chernobyl Nuclear Power Plant catastrophe). Therefore, purely temporal change‐points may be tested in time series of appropriate public health indicators (e.g. mortality, morbidity, sex ratio at birth). If, in addition, the spatial contamination is strong and variable enough and can be identified with sufficient precision at the level of regional units (e.g. districts), then a spatial‐temporal approach makes sense. This essentially means that a global time trend model is adjusted for region‐specific trend functions, allowing for local or global temporal jumps or broken sticks (change‐points) at certain points in time. The local jump heights may be tested for associations with local exposure (exposure–response relation), and all other characteristics in the data that vary with locality and in time are automatically accounted for, thus minimizing confounding. Spatial‐temporal approaches may help to strengthen the evidence of possible causal relationships. As an example, the human sex ratio at birth in several European countries before and after the Chernobyl Nuclear Power Plant accident was investigated. A long‐term chronic impact of radioactive fallout on the secondary sex ratio has been found. Copyright © 2008 John Wiley & Sons, Ltd.